文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > IT/计算机 > Issues in the Design of Relational Model Management Systems
侵权投诉

Issues in the Design of Relational Model Management Systems

关于模型集成方面比较早引用较多的英文文章,上次在网上找到一个英文网站可以下载,但是要收费,共享出来,大家有需要的可以看看

Issues in the Design of Relational Model Management Systems

397

INTRODUCTION The growing use of decision models in decision support systems (DSSs)1'14'15'22'30 and the successful implementation of database management systems (DBMSs) have given rise to suggestions that (1) decision models, like data, are an important information resource that should be managed effectively, (2) a discipline of model management be developed to help DSS users to make more effective use of models, and (3) model management systems be developed to implement important model management concepts.17'18'24'25'30,31'33 Examples of such models are mathematical programming models for production and distribution scheduling, Monte Carlo simulations of transportation networks, and deterministic simulations of the financial structure of an enterprise. Although no model management systems now exist, it can be argued that the systems that support such financial planning languages as EMPIRE13 and IFPS19'23 are forerunners of the model management systems that will eventually be implemented, just as the file processing systems of the 1960s were forerunners of DBMSs. One of the frameworks proposed for model management is a relational framework: it is suggested that a model bank may be viewed as a set of (virtual) relations with input and output attributes and functional dependencies between them, just as a database may be viewed as a set of relations with key and content attributes.4'5'6'7'8'10 Two advantages are offered for such a perspective. First, since some of the operations performed with models have counterparts in relational data management, important concepts in the latter area may be useful in the former. For example, the execution of a model is similar to selection or restriction of a data relation. Similarly, whenever several models are used to respond to a query such that the outputs of some models are inputs to others, an operation similar to a join is performed. Clearly there are substantial differences between relational data and model management at the level of system implementation and program execution, but we are concerned here only with the user view of models, not with the procedures for model execution. The second possible advantage is that it may eventually be possible to synthesize the relational frameworks for data and model management to produce a single relational framework for information management in DSSs.8 Three principal

issues have been identified for relational model management: 1. Model bank organization. Should a model bank be viewed by a DSS user as a single"universal" model or as a set of separate models? If the latter, what are the criteria for partitioning the universal model? Are there

normal forms for relational model banks as there are for relational databases? 2. Relational completeness of model query languages. What operations should be specified by a query language for relational model management? How do they differ from those of relational data management? What are the linguistic properties (e.g., position on the Chomsky hierarchy) of a relationally complete model query language? 3. Implementation issues. What are the issues with regard to security, integrity, query optimization, and the implementation of joins in relational model management? With regard to the latter, if an output of one model is an input to another and vice versa, how is a consistent solution for the model bank to be found? These three issues will be examined in the following three sections, after which we will discuss the broader issue of model management in DSSs. MODEL BANK ORGANIZATION The question of whether several interrelated operations (such as preparing a sales forecast and then simulating the impact of the resulting sales volume on production processes) should be performed by one model or by several models has been raised in the model management literature. It has been suggested that models be decomposed by organizational function (finance, marketing, etc.),33 organizational level (the strategic, tactical, and operational levels),29 model solution technique (simulation, linear programming, etc.),3'9 or a combination of these.12 Another approach, similar to that taken in relational data management, consists of identifying anomalies that can arise in using models and then deriving normal forms that result in the elimination of these anomalies.6 The anomalies of interest in relational data management are update anomalies— problems that can arise in adding, deleting, or changing tuples in a relation. Such anomalies do not arise in relational model management, because the tuples in a model relation do not exist in stored form; rather, they are generated as needed in response to user queries. However, other anomalies, called processing anomalies, are found in model management, and they give rise to normal forms similar to, but not identical with, those of relational data management. There are three types of processing anomaly, which lead to three types of normal forms, called the alpha, beta, and gamma forms. These anomalies are 1. Input anomalies.—An input anomaly occurs whenever a user, in order to require that an output of a model be

Issues in the Design of Relational Model Management Systems

上一页第3页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email